Kamil Potrec on authentication for AI agents

AI Engineer18m 34s
0 comments · 0 votesOpen discussionClose discussion
Sign in to join the discussion

    Kamil Potrec describes how agents use personal tokens, model API keys and service accounts, and why these patterns make it hard to determine who authorized an action. He uses an invoicing example to illustrate the attribution problem.

    He outlines requirements for agent identities, delegated authorityAn agent permission boundary defines and enforces which resources and actions an automated agent is allowed to use., short-lived credentials, federation with legacy systems, policy and audit. Potrec discusses the difficulty of applying these controls consistently across many services and agent types.

    Potrec compares SDKs with gateways that keep credentials outside the agent runtime and enforce policy on requests. He also describes gateway limitations, including latency, network bypasses, local execution and operational complexity, before introducing Keycard’s commercial approach.

    Original YouTube thumbnailWatch on YouTube

    Video topic coverage

    • Agents & workflows100
    • Coding & tools85
    • Models & learning0
    • Infrastructure & compute60
    • Research & evaluation15
    • Safety & security100
    • Business & productivity40
    • Creative & media0
    • Robotics & perception0
    • Society & policy0

    Share this page

    Kamil Potrec in a blue T-shirt beside the white and blue words Agent Authentication on black. With WWW.ARTIFICIAL-INTELLIGENCE.VIDEO, 28 October 2026 and duration 18m 34s.

    Kamil Potrec explains how distinct agent identities and scoped credentials can support safer authorization.